At SEMICON China, Wally Rhines gave one of the keynotes on the opening day. It was an update on his presentation Merger Mania that I covered a couple of years ago when Wally gave the keynote at that year's GSA Silicon Summit. Wally Rhines has been CEO of Mentor since 1993 (well, Mentor is a subsidiary of Siemens these days, but Wally still runs it). Prior to that his career was at Texas Instruments, where he ended up running their entire semiconductor business. Wally is clearly proud that TI looks to be the most profitable semiconductor company last year, with earnings over 40% of revenue. This presentation was partly an investigation into why. Acquisitions Wally's presentations often take some piece of received wisdom, such as the EDA industry is consolidating, and showing that it is wrong with data. This time he started off with the received wisdom that the number of semiconductor acquisitions is dramatically up. The average number of acquisitions from 2010 to 2016 was 25. In 2017, there were just 15, dramatically down. Having said that, the value of the acquisitions is up significantly, although his graph still included Broadcom's acquisition of Qualcomm, which had been blocked by CFIUS just a day before SEMICON China opened. But the last couple of years still included Avago/Broadcom, Intel/Altera, Softbank/Arm, NXP/Freescale, Qualcomm/NXP (probably), and more. Perhaps more surprising, the semiconductor industry has been deconsolidating (if that's not a word, it should be). The 50 largest semiconductor companies market share has gradually decreased 15 points in the decade to 2014, but that might have changed with the recent mergers. Another thing that might be ending the deconsolidation is the sudden ramp of fabless startups. There are some in the US, driven mostly by the interest in deep learning (but not entirely). But if you think that the center of fabless semiconductor startups is Silicon Valley, you need to get out more. Out of the country, that is. China is where the action is, with an accelerating ramp of new "IC Design Enterprises". Economies of Scale Next, Wally cast his critical eye on whether acquisitions are driven by economies of scale, since it is received wisdom that semiconductor profitability correlates with size, since semiconductor is a mass production business with "obvious" returns to scale. Except Wally had data, and showed that profitability doesn't correlate with scale (being Wally, he had the actual number, that over five years the correlation coefficient R 2 is 0.0706). If you look at the most profitable semiconductor companies (over $200M) in 2016, the last year where data is readily available, you find that the most profitable companies are not the big ones you might expect. A lot of analog. A lot of specialist foundries. So a lot of non-leading-edge processes. Wally took a more critical look at acquisitions. After all, the biggest companies got big through acquisitions, surely? Samsung has grown its market share from 10.2% to 12.1% over 2011 to 2016...but not from acquisitions Intel's market share has stayed flat (15.2% to 15.6%) despite a lot of acqusitions (Axxia, Lantiq, Altera, McAffee, Infineon Mobile) TSMC has almost doubled its market share from 4.5% to 8.1% with no acqusitions Qualcomm went from 3% up to 5.4% and down to 4.2%. It's gain and loss is driven by the underlying wireless market, not acqusitions Broadcom/Avago is the one top 5 semiconductor company that grew by acqusition, from 0.7% in 2011 to 4.2% in 2016 Specialization If doing acquisitions to get big doesn't seem to drive profitability, what does? Wally took a look at how companies specialized. In particular, companies that used acquisitions and divestitures to specialize, usually increased earnings percent. NXP increasingly focused on automotive and security...and their earnings percentage has grown as a result. TI has focused on analog and its earnings have gone from 10% (that Wally said would have been a "great year" when he was running TI semiconductor) to 40%. Broadcom+LSI+Avago has focused on datacenter, networking, and RF wireless, doing some acquisitions like LSI, but also then divesting some of the non-core assets such as selling Axxia to Intel. Its profitability is also close to 40% And a counterexample. Intel used to be one of the most focused semiconductor companies of all, pretty much microprocessors and not a lot else. Its recent acquisitions have diversified away, but its profitability has been in the 20-30% range for nearly 30 years. Of course, its revenues have grown, and so its profits are up. Until the recent runup in DRAM prices, it was the clear #1 in revenue, but in the middle of the pack in profitability. R&D Spending and Transistor Cost Reduction Continue Wally wrapped up with a little more data. Despite predictions that consolidation (which apparently isn't happening anyway) would lead to lower R&D, in the semiconductor industry, R&D has remained remarkably constant at 14% for over 35 years, and with no sign of any imminent change. On the other hand, despite the predictions about the end of Moore's Law, transistor cost reduction continues apace, based on the number of transistors shipped by the industry. Conclusions The final fact-based conclusions, which don't really fit the usual narrative but depend on actual data, are: Consolidation of the semiconductor industry is limited Profitability in the semiconductor industry is driven by specialization, not economies of scale The semiconductor industry will not consolidate into a few large companies Semiconductor R&D continues to grow and is maintaining historical levels as a percent of revenue So next time someone tells you that the way for a semiconductor company to improve its margins is through acquisition, and so eventually there will be just a handful left standing, you now have some data that points to the opposite conclusion. Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧